Abstract

Across linguistic theories, human language structures are represented by graphs (e.g., Chomsky, 1957, Tesnière, 1959, Chomsky, 1995). Much research has focused on the mapping between such graphs and the actual sequences expressing utterances, but less attention has been paid to the shapes that the graphs themselves take: their topologies. A current hypothesis argues that the structures in human language are primarily shaped by language production (Bock, 1982, Bock & Warren, 1985, MacDonald, 1999, Jäger & Rosenbach, 2008, MacDonald, 2013). Utterances are planned in an incremental manner: successively incorporating chunks—either single words or larger units—into partial syntactic structures (Bock, 1982, Bock & Warren, 1985, Bock, 1987, Levelt, 1989, Bock & Levelt, 1994, Ferreira & Dell,2000). Incremental construction should constrain the plausible probability distributions of syntactic structures. I will show that the topologies of actual syntactic graphs exhibit the precise deviation from randomness that incremental construction predicts. This is a previously unknown universal regularity of human languages: Syntactic structures are constrained to a predictable topological distribution—that generated by sublinear preferential attachment (Krapivsky et al., 2000, Barabasi & Posfai, 2016)—constant for all 124 languages studied, across language families and modalities (spoken, written, and signed). It supports the hypothesis that syntactic structures are mainly shaped by language production. Furthermore, it demonstrates how the observed efficiency of languages might arise without any optimization process. This finding implicitly defines a data-free universal prior distribution for parse structures, with possible applications in language technologies.